Deep Learning-Based Short-Term Heavy Rainfall Model

This study presents an Att-LSTM model based on a humidity threshold attention mechanism, applied to rainfall prediction in Wuhu, Anhui Province, with the aim of improving the accuracy of short-term heavy rainfall forecasts. The model combines the time series modeling capabilities of Long Short-Term Memory (LSTM) networks and introduces a humidity threshold-guided attention mechanism, which enhances the model's focus on critical time steps when humidity values exceed a preset threshold through a weighting mechanism. Comparative experimental results show that the attention model incorporating the humidity threshold exhibits higher accuracy and better stability in rainfall period identification and heavy rainfall prediction compared to the traditional LSTM model. The humidity threshold-based attention mechanism allows the model to adapt more flexibly to sudden weather events, ensuring that rainfall predictions align more closely with actual weather patterns. The findings indicate that the humidity threshold mechanism can effectively enhance the performance of rainfall prediction models, providing new technical support for improving meteorological forecasting accuracy and disaster prevention.